import os
import umap
import numpy as np
import seaborn as sns
from   mnist import MNIST
import matplotlib.pyplot as plt
import matplotlib.colors as mcolors
from   sklearn.cluster import KMeans
from   sklearn.datasets import fetch_openml
from   sklearn.model_selection import train_test_split
from   sklearn.metrics import adjusted_mutual_info_score

WARNING:tensorflow:From c:\Users\Carlos\anaconda3\envs\Intro_CD_23_24\Lib\site-packages\keras\src\losses.py:2976: The name tf.losses.sparse_softmax_cross_entropy is deprecated. Please use tf.compat.v1.losses.sparse_softmax_cross_entropy instead.

#%matplotlib inline

sns.set_theme(style='white', context='poster')

#En la carperta Fashion-MNIST están descargado los datos
mndata = MNIST('data/fashion')

X_train, y_train = mndata.load_training()
X_test,  y_test  = mndata.load_testing()

data = np.array(np.vstack([X_train, X_test]), dtype=np.float64) / 255.0
target = np.hstack([y_train, y_test])
classes = [
    'T-shirt/top',
    'Trouser',
    'Pullover',
    'Dress',
    'Coat',
    'Sandal',
    'Shirt',
    'Sneaker',
    'Bag',
    'Ankle boot']

embedding1 = umap.UMAP(n_neighbors=10).fit_transform(data)

fig, ax = plt.subplots(1, figsize=(10, 6))
plt.scatter(*embedding1.T, s=0.3, c=target, cmap='Spectral', alpha=1.0)
plt.setp(ax, xticks=[], yticks=[])
cbar = plt.colorbar(boundaries=np.arange(11)-0.5)
cbar.set_ticks(np.arange(10))
cbar.set_ticklabels(classes)
plt.title('Fashion MNIST Embedded via UMAP (10 vecinos)');

mapper1 = umap.UMAP(densmap=True, n_neighbors=10).fit(data)

fig, ax = plt.subplots(1, figsize=(10, 6))
plt.scatter(*mapper1.embedding_.T, s=0.3, c=np.array(target), cmap='Spectral', alpha=1.0)
plt.setp(ax, xticks=[], yticks=[])
cbar = plt.colorbar(boundaries=np.arange(11)-0.5)
cbar.set_ticks(np.arange(10))
cbar.set_ticklabels(classes)
plt.title('Fashion MNIST Embedded via DensMAP (10 vecinos)');

embedding2 = umap.UMAP(n_neighbors=15).fit_transform(data, y=target)

fig, ax = plt.subplots(1, figsize=(10, 6))
plt.scatter(*embedding2.T, s=0.1, c=target, cmap='Spectral', alpha=1.0)
plt.setp(ax, xticks=[], yticks=[])
cbar = plt.colorbar(boundaries=np.arange(11)-0.5)
cbar.set_ticks(np.arange(10))
cbar.set_ticklabels(classes)
plt.title('Fashion MNIST Embedded via UMAP (con etiquetas)');

mapper2 = umap.UMAP(densmap=True, n_neighbors=15).fit(data, np.array(target))

fig, ax = plt.subplots(1, figsize=(10, 6))
plt.scatter(*mapper2.embedding_.T, s=0.3, c=np.array(target), cmap='Spectral', alpha=1.0)
plt.setp(ax, xticks=[], yticks=[])
cbar = plt.colorbar(boundaries=np.arange(11)-0.5)
cbar.set_ticks(np.arange(10))
cbar.set_ticklabels(classes)
plt.title('Fashion MNIST Train Digits Embedded via DensMAP (con etiquetas)')

Text(0.5, 1.0, 'Fashion MNIST Train Digits Embedded via DensMAP (con etiquetas)')

embedding3 = umap.UMAP(n_neighbors=15).fit_transform(X_train, y=y_train)

fig, ax = plt.subplots(1, figsize=(10, 6))
plt.scatter(*embedding3.T, s=0.3, c=y_train, cmap='Spectral', alpha=1.0)
plt.setp(ax, xticks=[], yticks=[])
cbar = plt.colorbar(boundaries=np.arange(11)-0.5)
cbar.set_ticks(np.arange(10))
cbar.set_ticklabels(classes)
plt.title('Fashion MNIST Embedded via UMAP (entrenamiento con etiquetas)');

umap_model = umap.UMAP(n_neighbors=15).fit(X_train, y=y_train)
embedding_test = umap_model.transform(X_test)

fig, ax = plt.subplots(1, figsize=(10, 6))
plt.scatter(*embedding_test.T, s=0.3, c=y_test, cmap='Spectral', alpha=1.0)
plt.setp(ax, xticks=[], yticks=[])
cbar = plt.colorbar(boundaries=np.arange(11)-0.5)
cbar.set_ticks(np.arange(10))
cbar.set_ticklabels(classes)
plt.title('Fashion MNIST Embedded via UMAP (prueba sin etiquetas)')

Text(0.5, 1.0, 'Fashion MNIST Embedded via UMAP (prueba sin etiquetas)')

# Create the directory if it doesn't exist
os.makedirs('data/plots', exist_ok=True)


# Define the range of parameters
n_neighbors_range = range(4, 11, 1)
min_dist_range = (0.1, 0.5, 0.90)

# Initialize variables to store the best parameters and score
best_score = -1
best_n_neighbors = None
best_min_dist = None


# Initialize lists to store the scores and parameter combinations
scores = []
param_combinations = []


# Iterate over the parameter combinations
for n_neighbors in n_neighbors_range:
    for min_dist in min_dist_range:
        # Perform UMAP embedding with the current parameters
        umap_model = umap.UMAP(n_neighbors=n_neighbors, min_dist=min_dist)
        embedding = umap_model.fit_transform(data)
        
        # Perform K-Means clustering on the embedded data
        kmeans = KMeans(n_clusters=10)
        labels_prueba = kmeans.fit_predict(embedding)
        
        # Calculate the adjusted mutual information score
        score = adjusted_mutual_info_score(target, labels_prueba)
        
        # Store the score and parameter combination
        scores.append(score)
        param_combinations.append((n_neighbors, min_dist))
        
        # Update the best score and parameters if necessary
        if score > best_score:
            best_score = score
            best_n_neighbors = n_neighbors
            best_min_dist = min_dist
        
        
        
        fig, ax = plt.subplots(1, figsize=(10, 6))
        plt.scatter(embedding[:, 0], embedding[:, 1], s=0.3, c=labels_prueba, cmap='Spectral', alpha=0.8)
        plt.title(f"UMAP Scatterplot (vecinos={n_neighbors}, mindist={min_dist})", fontsize=18)
        plt.setp(ax, xticks=[], yticks=[])
        cbar = plt.colorbar(boundaries=np.arange(11)-0.5)
        cbar.set_ticks(np.arange(10))
        cbar.set_ticklabels(classes)
        cbar.ax.tick_params(labelsize=12)
        plt.xlabel('', fontsize=14)
        plt.ylabel('', fontsize=14)
        plt.xticks(size=12)
        plt.yticks(size=12)
        
        # Save the plot
        plot_name = f"New2_UMAP Scatterplot (n_neighbors={n_neighbors}, min_dist={min_dist})"
        plt.savefig(f"data/plots/{plot_name}.png")
        
        # Show the plot
        plt.show()

c:\Users\Carlos\anaconda3\envs\Intro_CD_23_24\Lib\site-packages\sklearn\cluster\_kmeans.py:1412: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning
  super()._check_params_vs_input(X, default_n_init=10)

c:\Users\Carlos\anaconda3\envs\Intro_CD_23_24\Lib\site-packages\sklearn\cluster\_kmeans.py:1412: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning
  super()._check_params_vs_input(X, default_n_init=10)

c:\Users\Carlos\anaconda3\envs\Intro_CD_23_24\Lib\site-packages\sklearn\cluster\_kmeans.py:1412: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning
  super()._check_params_vs_input(X, default_n_init=10)

c:\Users\Carlos\anaconda3\envs\Intro_CD_23_24\Lib\site-packages\sklearn\cluster\_kmeans.py:1412: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning
  super()._check_params_vs_input(X, default_n_init=10)

c:\Users\Carlos\anaconda3\envs\Intro_CD_23_24\Lib\site-packages\sklearn\cluster\_kmeans.py:1412: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning
  super()._check_params_vs_input(X, default_n_init=10)

c:\Users\Carlos\anaconda3\envs\Intro_CD_23_24\Lib\site-packages\sklearn\cluster\_kmeans.py:1412: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning
  super()._check_params_vs_input(X, default_n_init=10)

c:\Users\Carlos\anaconda3\envs\Intro_CD_23_24\Lib\site-packages\sklearn\cluster\_kmeans.py:1412: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning
  super()._check_params_vs_input(X, default_n_init=10)

c:\Users\Carlos\anaconda3\envs\Intro_CD_23_24\Lib\site-packages\sklearn\cluster\_kmeans.py:1412: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning
  super()._check_params_vs_input(X, default_n_init=10)

c:\Users\Carlos\anaconda3\envs\Intro_CD_23_24\Lib\site-packages\sklearn\cluster\_kmeans.py:1412: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning
  super()._check_params_vs_input(X, default_n_init=10)

c:\Users\Carlos\anaconda3\envs\Intro_CD_23_24\Lib\site-packages\sklearn\cluster\_kmeans.py:1412: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning
  super()._check_params_vs_input(X, default_n_init=10)

plt.figure(figsize=(10, 6))
plt.plot(scores, linestyle='--', color='blue', marker='o', markersize=7)
plt.xlabel('Parameter Combination')
plt.ylabel('Adjusted Mutual Information Score')
plt.title('Adjusted Mutual Information Score vs Parameter Combination')
plt.xticks(range(len(param_combinations)), param_combinations, rotation=45, size=12)
plt.show()

# Perform UMAP embedding with the best parameters
best_umap_model = umap.UMAP(n_neighbors=best_n_neighbors, min_dist=best_min_dist)
best_embedding = best_umap_model.fit_transform(data)

# Perform K-Means clustering on the embedded data
kmeans = KMeans(n_clusters=10)
best_labels = kmeans.fit_predict(best_embedding)
best_labels_cat = [str(i) for i in best_labels]

c:\Users\Carlos\anaconda3\envs\Intro_CD_23_24\Lib\site-packages\sklearn\cluster\_kmeans.py:1412: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning
  super()._check_params_vs_input(X, default_n_init=10)

fig, ax = plt.subplots(1, figsize=(10, 6))
plt.scatter(best_embedding[:, 0], best_embedding[:, 1], s=0.3, c=best_labels, cmap='Spectral', alpha=0.8)
plt.title(f"Real Best UMAP Scatterplot ({best_n_neighbors},{best_min_dist})", fontsize=14)
plt.setp(ax, xticks=[], yticks=[])
cbar = plt.colorbar(boundaries=np.arange(11)-0.5)
cbar.set_ticks(np.arange(10))
cbar.set_ticklabels(classes)
cbar.ax.tick_params(labelsize=10)
plt.xlabel('UMAP Dimension 1', fontsize=14)
plt.ylabel('UMAP Dimension 2', fontsize=14)
plt.xticks(size=12)
plt.yticks(size=12)
plt.show()

# Obtener los parámetros y scores de todos los modelos
param_scores = [(n_neighbors, min_dist, score) for (n_neighbors, min_dist), score in zip(param_combinations, scores)]

# Ordenar los modelos por score de mayor a menor
param_scores.sort(key=lambda x: x[2], reverse=True)

# Obtener los valores de los parámetros y scores
param_values = [f"n_neighbors={n_neighbors}, min_dist={min_dist}" for n_neighbors, min_dist, _ in param_scores]
score_values = [score for _, _, score in param_scores]

# Crear el gráfico de barras
plt.figure(figsize=(12, 6))
plt.bar(param_values, score_values)
plt.xticks(rotation=90)
plt.xlabel('Modelos')
plt.ylabel('Score')
plt.title('Scores de los modelos probados')
plt.show()

n_neighbors=9
min_dist=0.1

real_best_umap_model = umap.UMAP(n_neighbors=n_neighbors, min_dist=min_dist)
embedding = real_best_umap_model.fit_transform(data, y=target)
        
# Perform K-Means clustering on the embedded data
kmeans = KMeans(n_clusters=10)
labels = kmeans.fit_predict(embedding)

c:\Users\Carlos\anaconda3\envs\Intro_CD_23_24\Lib\site-packages\sklearn\cluster\_kmeans.py:1412: FutureWarning: The default value of `n_init` will change from 10 to 'auto' in 1.4. Set the value of `n_init` explicitly to suppress the warning
  super()._check_params_vs_input(X, default_n_init=10)

fig, ax = plt.subplots(1, figsize=(10, 6))
plt.scatter(embedding[:, 0], embedding[:, 1], s=0.3, c=labels, cmap='Spectral', alpha=0.8)
plt.title(f"Real Best UMAP Scatterplot ({n_neighbors},{min_dist})", fontsize=14)
plt.setp(ax, xticks=[], yticks=[])
cbar = plt.colorbar(boundaries=np.arange(11)-0.5)
cbar.set_ticks(np.arange(10))
cbar.set_ticklabels(classes)
cbar.ax.tick_params(labelsize=10)
plt.xlabel('UMAP Dimension 1', fontsize=14)
plt.ylabel('UMAP Dimension 2', fontsize=14)
plt.xticks(size=12)
plt.yticks(size=12)
plt.show()

Ejercicios UMAP ¶

UMAP en Fashion MNIST ¶

Utilizando etiquetas para separar clases (Supervised UMAP)¶

Entrenamiento con etiquetas y embedding con test no etiquetado (Metric Learning with UMAP)¶

Agrupamiento (clustering)¶

Ejercicios UMAP¶

UMAP en Fashion MNIST¶

Utilizando etiquetas para separar clases (Supervised UMAP)¶

Entrenamiento con etiquetas y embedding con test no etiquetado (Metric Learning with UMAP)¶

Agrupamiento (clustering)¶

Ejercicios UMAP ¶

UMAP en Fashion MNIST ¶